Schedules: the agent shows up by itself
Lesson 12 of 22 in the course “AI Agents for Life”. Free, with animations and a quiz.
From chat to autopilot
Until now you’ve launched the agent yourself: opened a chat, pasted the brief, got an answer. Forget, and nothing happens. The next step is a schedule. The brief is set up once and the agent shows up by itself: a briefing at 7:30, a digest on Friday, a weekly review on Sunday. That’s the “autopilot” in this level’s name. It works while you sleep or sit in class. But autopilot has a cost: you don’t see every run, so the rules in the brief need to be stricter than in a normal chat.
Where it already exists
Schedules are already built into the big assistants. In January 2025 ChatGPT started rolling out scheduled tasks: you write “every weekday at 7:30 send me a briefing,” and it runs the brief by itself and sends a notification. In June 2025 Google added scheduled actions to Gemini, first for paid plans. There’s a limit on active tasks, so five separate digests are better merged into one. And for linking apps together there are Zapier and n8n, which we’ll cover in the next lesson.
What to put on a schedule
Not everything is worth automating. Tasks that repeat and where regularity matters work well. A morning briefing every day. A digest on a topic you’re studying three times a week. Watching once a day: are there concert tickets, a new job, has a price dropped. And a weekly review on Sunday evening. One-off, important decisions work badly: choosing a university, a big purchase. They need you, not autopilot. Liza started with a single briefing and only added a digest two weeks later.
The quiet rule
Autopilot’s biggest mistake is noise. A watcher agent writes “nothing new” every day, and after a week you stop opening its messages. Then on the day the job actually appears, you miss it. The fix is a quiet rule right in the brief: “only message me when there’s something new; otherwise stay quiet.” The second rule: a short format. A notification should take 10 seconds to read. Want more detail? You’ll ask. One useful message a week beats seven empty ones.
Try it: the weekly review
Put the most useful habit agent on a schedule: the weekly review. In ChatGPT, write the brief below and add “every Sunday at 20:00” at the start, and the assistant creates the task itself. In Gemini you do the same with scheduled actions, if your plan has them. No scheduling? Set a repeating reminder on your phone and paste the brief by hand: that works too. Liza has done this for two months: five lines on Sundays help her remember what it’s all for.
Check yourself
- What should change in the brief?
Correct answer: Add a quiet rule: message only when there’s something new. Empty messages train you to ignore them, and the important one gets lost. With a quiet rule, every notification means something. - Which of these is worth scheduling?
Correct answer: Checking ticket prices once a day. A repeated check where regularity matters is the perfect scheduled task. - What’s the best move?
Correct answer: Merge them into one digest with sections. One digest doesn’t hit the active task limit and is read in one go instead of five notifications. - Why are scheduled briefs stricter?
Correct answer: You don’t see every run and can’t correct it on the fly. In a chat you see a mistake right away and fix it. On autopilot, the mistake repeats every day until you notice.
← Study and career: a tutor and a recruiter “If → then”: autopilot without code →